Parameter neighborhoods that survive a shift

Settings that stay profitable as conditions change

Parameter choice is framed as finding settings that stay profitable while limiting equity drawdown as conditions change, not as maximizing closed cumulative profit.

Optimal-parameter-shift is the change, from one sample window to the next, in the settings that look best on a completed-trade score. Commodity systems are said to be able to remain profitable for a given market after shifts as large as 50%. The stated search target is an operating point profitable enough that shifts of 25% or more still avoid an annual loss for any commodity under test.

How the completed-trade score is kept

The evaluation protocol uses one-contract trading, 50 to 100 currency units of commission and slippage per closed trade, no entries on locked-limit days, and at least six years of prices.

Open trades are dropped from the score unless they have already exceeded their peak equity. Intra-trade swings are accumulated as peak-to-trough-maximum-equity-drawdown until a new equity high, after which the accumulator resets.

Wide scan and narrow scan

Optimization is described as a wide-scan, then a narrow-scan, continued until a 10% move in any parameter still meets an annual 2.5-to-1 profit-to-drawdown-screen. If two neighborhoods score almost the same on that screen, the lower peak-to-trough-maximum-equity-drawdown is preferred.

Sector values and equity spikes

Average-sector-value is the mean of a nine-cell grid around a candidate center. The sector-central-value is to be used only when neighboring cells keep closed profit from falling sharply.

A candidate center is treated as an equity-spike and not used for trading if it sits 10% or more above any neighbor. A related check is whether more than 10% of closed profit comes from a single trade. If it does, that trade is removed and the score is recomputed.

The moving-average band example

In the moving-average-plus-band example, a close through the upper edge of a percentage-price-band is a buy, a close through the lower edge is a sell, and the moving average is the exit stop. A 10% change in lookback or band width is expected to change closed profit by no more than 10% if the center is robust.

The first-pass moving-average wide-scan covers lookbacks from 5 to 35 days and band widths from 2.4% to 4.5%. Later neighborhood comparisons of drawdown and the profit-to-drawdown-screen are used to pick among several candidate centers rather than taking the single largest closed-profit cell.